Motivated by pervasive biomedical data, we propose a unified mediation analysis approach to complex data of mixed types, including continuous, categorical, count variables. We invoke copula models to specify joint distributions of outcome variables, mediators and exposure variables of interest in the context of generalized linear models. We develop inference procedures to evaluate casual pathways in both aspects of parameter estimation and hypothesis testing for direct and/or indirect effects of the exposure variable on outcome variables. Our proposed method also enables us to identify important mediators through which exposure variables have indirect effects. We examine necessary model assumptions for the identifiability of casual effects and establish asymptotic properties for the proposed method. We compare the performance of the proposed method with other existing methods using simulation studies. We apply the proposed method to an analysis of a real biomedical dataset.